A Precision-First Whitespace Restoration Pipeline for Noisy Urdu Tafseer Text

Urdu Qur’anic commentary (tafseer) reaches computational tools through lossy conversions (desktop-publishing exports, OCR, PDF text extraction) that damage word boundaries. We summarize a confidence-gated, space-only pipeline that restores missing spaces in a converted Urdu tafseer while being verifiably unable to alter any letter of the text. The pipeline proposes only space-insertion edits, scores each with corpus-internal lexical evidence, holds Qur’anic Arabic and other protected regions out of automatic correction, and verifies a character-preservation invariant. On a 1,000-row silver-labeled sample, its auto-accept operating point reaches 96.9% strict precision on decisive cases, against 30.5% for an ungated splitter and 39.7% for a frequency-only baseline, with zero character-altering edits. We argue that this conservative design is an appropriate default for religious and scholarly text collections, and we set out its limitations.